Guides

Regression workflow

1. Train or define any black-box predictor f(x).
2. Wrap it in SelfCalibratingConformalPredictor.
3. Calibrate on an independent calibration set.
4. Use predict_point, predict_venn_abers, and predict_interval on new data.
5. Use check_coverage and plot to inspect empirical behavior on held-out data.

Quantile-loss workflow

1. Choose a conformity score S(x, y), typically |y - mu(x)|.
2. Train a predictor for the (1 - alpha) quantile of that score.
3. Wrap the quantile predictor in VennAbersQuantileConformalPredictor.
4. Calibrate using isotonic quantile calibration.
5. Form intervals from the score level set { y : S(x, y) <= threshold(x, y) }.
6. Inspect both marginal coverage and threshold calibration with check_threshold_calibration.

Customization hooks

• Provide any callable or .predict model as the base predictor.
• Swap in a custom calibrator with the same signature as the built-ins.
• Provide a custom conformity score on either the regression or quantile workflow.
• Use the advanced notebook when you want custom-score examples instead of the default absolute-residual construction.